Machine Learning-Based Approach for Automatic Ion Implanter Monitoring

Ion implanter is a bottleneck equipment in the semiconductor device fabrication due to its expensiveness. Ion implantation is a doping method that introduces impurities into a semiconductor wafer. Since the wafer is not reusable if any fault occurs, real-time fault detection to minimize the possible downtime of the ion implanter is extremely important. In this paper, we firstly investigate seven machine learning-based methods. Then, we perform three models: decision tree, random forest, and AdaBoost for recipe classification in detecting anomalies. These three models are relatively comprehensible compared with neural network-based ones and might be useful for providing knowledge to improve the manufacturing process. Experimental results show that the AdaBoost is the best one and can be practically useful for automatic monitoring of the ion implanter.